diff options
Diffstat (limited to 'modules/swinir_model.py')
| -rw-r--r-- | modules/swinir_model.py | 49 |
1 files changed, 34 insertions, 15 deletions
diff --git a/modules/swinir_model.py b/modules/swinir_model.py index 9bd454c6..facd262d 100644 --- a/modules/swinir_model.py +++ b/modules/swinir_model.py @@ -7,10 +7,10 @@ from PIL import Image from basicsr.utils.download_util import load_file_from_url from tqdm import tqdm -from modules import modelloader -from modules.paths import models_path -from modules.shared import cmd_opts, opts, device +from modules import modelloader, devices +from modules.shared import cmd_opts, opts from modules.swinir_model_arch import SwinIR as net +from modules.swinir_model_arch_v2 import Swin2SR as net2 from modules.upscaler import Upscaler, UpscalerData precision_scope = ( @@ -25,7 +25,6 @@ class UpscalerSwinIR(Upscaler): "/003_realSR_BSRGAN_DFOWMFC_s64w8_SwinIR" \ "-L_x4_GAN.pth " self.model_name = "SwinIR 4x" - self.model_path = os.path.join(models_path, self.name) self.user_path = dirname super().__init__() scalers = [] @@ -43,7 +42,7 @@ class UpscalerSwinIR(Upscaler): model = self.load_model(model_file) if model is None: return img - model = model.to(device) + model = model.to(devices.device_swinir) img = upscale(img, model) try: torch.cuda.empty_cache() @@ -59,22 +58,42 @@ class UpscalerSwinIR(Upscaler): filename = path if filename is None or not os.path.exists(filename): return None - model = net( + if filename.endswith(".v2.pth"): + model = net2( upscale=scale, in_chans=3, img_size=64, window_size=8, img_range=1.0, - depths=[6, 6, 6, 6, 6, 6, 6, 6, 6], - embed_dim=240, - num_heads=[8, 8, 8, 8, 8, 8, 8, 8, 8], + depths=[6, 6, 6, 6, 6, 6], + embed_dim=180, + num_heads=[6, 6, 6, 6, 6, 6], mlp_ratio=2, upsampler="nearest+conv", - resi_connection="3conv", - ) + resi_connection="1conv", + ) + params = None + else: + model = net( + upscale=scale, + in_chans=3, + img_size=64, + window_size=8, + img_range=1.0, + depths=[6, 6, 6, 6, 6, 6, 6, 6, 6], + embed_dim=240, + num_heads=[8, 8, 8, 8, 8, 8, 8, 8, 8], + mlp_ratio=2, + upsampler="nearest+conv", + resi_connection="3conv", + ) + params = "params_ema" pretrained_model = torch.load(filename) - model.load_state_dict(pretrained_model["params_ema"], strict=True) + if params is not None: + model.load_state_dict(pretrained_model[params], strict=True) + else: + model.load_state_dict(pretrained_model, strict=True) if not cmd_opts.no_half: model = model.half() return model @@ -92,7 +111,7 @@ def upscale( img = img[:, :, ::-1] img = np.moveaxis(img, 2, 0) / 255 img = torch.from_numpy(img).float() - img = img.unsqueeze(0).to(device) + img = img.unsqueeze(0).to(devices.device_swinir) with torch.no_grad(), precision_scope("cuda"): _, _, h_old, w_old = img.size() h_pad = (h_old // window_size + 1) * window_size - h_old @@ -120,8 +139,8 @@ def inference(img, model, tile, tile_overlap, window_size, scale): stride = tile - tile_overlap h_idx_list = list(range(0, h - tile, stride)) + [h - tile] w_idx_list = list(range(0, w - tile, stride)) + [w - tile] - E = torch.zeros(b, c, h * sf, w * sf, dtype=torch.half, device=device).type_as(img) - W = torch.zeros_like(E, dtype=torch.half, device=device) + E = torch.zeros(b, c, h * sf, w * sf, dtype=torch.half, device=devices.device_swinir).type_as(img) + W = torch.zeros_like(E, dtype=torch.half, device=devices.device_swinir) with tqdm(total=len(h_idx_list) * len(w_idx_list), desc="SwinIR tiles") as pbar: for h_idx in h_idx_list: |